AI21, AIx, OpenAI and HuggingFace with Pen.el

1 · Shane Mulligan · Aug. 23, 2021, 9:11 a.m.
Summary I benchmark using several different LM engines in Pen.el. Benchmark 100 tokens AI21 vs AIx vs OpenAI AI21’s service is about as fast as OpenAI’s at about 6 seconds for 100 tokens, but AIx is significantly slower. You may speed up AIx’s generations by subscribing to the GPU plan. AI21 1 2 3 4 5 title: j1-jumbo include: Generic completion 50 tokens/1 prompt-version: 1 engine: AI21 J1-Jumbo max-tokens: 100 1 2 3 4 5 6 7 engine-title: AI21 J1-Jumbo lm-command: "ai21-complete....